Getting Started

Write your first IntentFlow workflow in 10 minutes

1Create a Workflow File

Create a new file called my-workflow.md and add the basic structure:

my-workflow.md 
# Workflow: Daily Weather Report

## Meta
version: 1.0
estimated_time: 5 minutes

## Context
Fetch weather data and create a simple summary report.

2Add Your First Step

Each step needs a Task that describes what to accomplish:

my-workflow.md (continued) 
## Step 1: Fetch Weather Data

### Task
Get the current weather for San Francisco from any weather API.
Save the temperature, conditions, and humidity.

### Save as
`/tmp/workflow/weather.json`

### Success criteria
- File contains temperature in Fahrenheit
- File contains current conditions (sunny, cloudy, etc.)
- File contains humidity percentage
💡 Key Insight

Notice how we describe what we want, not how to get it. The LLM will figure out which API to use and how to call it.

3Add More Steps

Steps are connected through files. The output of Step 1 becomes input for Step 2:

my-workflow.md (continued) 
## Step 2: Generate Report

### Task
Create a simple text report with:
- Current temperature and conditions
- A recommendation (umbrella, sunscreen, jacket)
- Tomorrow's outlook if available

### Flexibility
Format the report however you think is most readable.

### Save as
`/tmp/workflow/weather_report.txt`

### Success criteria
- File exists and is readable
- Contains all requested information

4Add Finalization

my-workflow.md (continued) 
## Finalization

### After completion
Display the contents of the report.

### If something went wrong
Show what data was retrieved and where the error occurred.

5Run Your Workflow

Simply provide the workflow to an LLM with tool access:

Please execute this workflow:

[paste your workflow here]

Or if the LLM has file access:

Please execute the workflow in my-workflow.md

The LLM will:

  1. Read and understand the workflow
  2. Execute each step in order
  3. Verify success criteria
  4. Handle any errors according to your instructions
  5. Provide the finalization summary

Key Concepts

Section Purpose
Task What to accomplish (required)
Save as Output file path — creates contract between steps
Success criteria How to verify step completion
Flexibility Where the LLM can improvise
If something goes wrong Fallback strategies

Common Mistakes

❌ Too vague
### Task
Do the data analysis.
✅ Better
### Task
Calculate the average, median, and standard deviation 
of the 'price' column. Group results by 'category'.
❌ Over-specified
### Task
Use requests library to call api.weather.gov endpoint
/points/37.7749,-122.4194 then parse the JSON response
using json.loads() and extract the 'temperature' key...
✅ Better
### Task
Get current weather for San Francisco.
Extract temperature and conditions.

Next Steps